The study addresses the question of how multiple languages are represented in the brain of polyglots. It explores whether there is a shared neural network for all languages or whether separate representations exist for each language. Understanding the neural mechanisms underlying multilingualism has significant implications for language acquisition, cognitive neuroscience, and educational practices. Previous research has yielded conflicting results, with some studies suggesting distinct neural substrates for each language and others supporting a shared network. This study aimed to clarify these discrepancies using a rigorous fMRI design and analysis, focusing on a large cohort of polyglots with diverse linguistic backgrounds and proficiency levels. The importance lies in resolving the debate regarding the neural architecture of multilingualism and understanding the neural flexibility associated with language learning and use in individuals who speak multiple languages fluently.

The literature review would cover previous studies on language representation in the brain, contrasting those that found distinct brain areas for different languages with those proposing a shared neural network. It would also discuss the existing literature on the role of language proficiency, language dominance, and age of acquisition in shaping neural representations. The review would emphasize the need for a study with a large sample size, diverse linguistic backgrounds, and a controlled experimental design to resolve inconsistencies in previous findings. Studies on bilingualism and multilingualism would be reviewed, highlighting the different methods and findings relating to neural activation patterns.

The study employed fMRI to measure brain activity in 34 polyglots while they listened to auditory stimuli in various languages. Participants listened to their native language (L1), three other familiar languages (L2, L3, L4) with varying proficiency levels, two unfamiliar related languages (URLs), two unfamiliar unrelated languages (UULs), and a control condition (Quilts). Language proficiency was assessed using self-reported measures of auditory and written comprehension and production. The age of acquisition was also recorded for each language. fMRI data were analyzed using linear mixed-effects models to investigate the relationship between brain activation patterns and language conditions, controlling for individual differences in language proficiency and age of acquisition. Language fROIs were defined by contrasts from a language localizer task. Robustness checks were performed using different materials (Bible vs. Alice in Wonderland), considering age and handedness, and across scanning runs. Data were also analyzed by excluding participants with low L1 proficiency, early balanced bilinguals, and those with errors in the selection of related languages. Additional analyses examined the responses in the right hemisphere homologue of the language network and the Multiple Demand (MD) network. The analyses examined the relationships between brain responses and language conditions, considering the factors of language proficiency, age of acquisition, and language relatedness. The study also controlled for potential confounds such as age and handedness.

The key finding was that all languages elicited a reliable response relative to the control condition in the left hemisphere language network. The response magnitudes for familiar languages (L1-L4) were greater than for unfamiliar languages. In the right hemisphere, the difference between familiar and unfamiliar languages was less pronounced. A mixed-effects model analysis showed a reliable effect of Condition (language) on brain activation, confirming that different language conditions elicited different patterns of neural activity. The response in the polyglots' language network was robust across scanning runs and to the materials used (Bible vs. Alice in Wonderland). The pattern of responses was largely independent of participants' age. Excluding participants with low native language proficiency, early balanced bilinguals, or errors in URL selection did not substantially alter the main findings. Analyses revealed a numerically lower response to the native language in polyglots than in monolingual controls, but this difference was not statistically significant (p = 0.063). The analysis in the MD network showed that familiar, related, and unrelated languages, and the Quilts control elicited above-baseline responses, whereas native language processing elicited a below-baseline response. Statistical comparisons of response magnitudes (Tables S4 and S5) showed significant differences between language conditions and the control in most language fROIs. The inter-language distance metric from Beaufils & Tomin (2020) was used to assess the relatedness of the unfamiliar languages used in the experiment (Table S6). The results from the LME model indicated that the condition response pattern was robust to the materials used and the participants' ages (Text S1 and S2). Figures S1-S15 provide detailed visualizations of the results across different conditions and participant subsets.

The findings support the hypothesis of a shared language network in polyglots, indicating that the brain represents multiple languages within a common neural substrate. The effect of language proficiency was clearly shown, with more proficient languages eliciting greater activation. However, the results also highlight the importance of language familiarity, as unfamiliar languages elicited less activation than familiar languages, even when considering the relatedness between languages. The numerically lower response to the native language in polyglots compared to controls is a notable finding that deserves further investigation. It suggests potential adaptation in native language processing in highly proficient multilingual individuals. The robustness of the findings across different analysis strategies increases confidence in the reliability and generalizability of the results. This study strengthens the evidence for neural plasticity and adaptation in response to multilingual experience.

This fMRI study provides compelling evidence for a shared neural network underlying multiple language processing in polyglots. The findings highlight the interaction between language proficiency and familiarity in shaping neural representations. Future research could investigate the longitudinal development of these neural representations and explore the influence of factors such as language dominance and context of language use on neural activation patterns. Further investigation into the observed numerical difference in native language processing between polyglots and controls is warranted.

The study's reliance on self-reported language proficiency might introduce some bias. Future studies could incorporate more objective measures of language proficiency. The cross-sectional nature of the study limits inferences about the developmental trajectory of language representations. Longitudinal studies are needed to address this limitation. The relatively small number of participants in some analyses might also affect the statistical power. Future studies should expand the sample size to achieve greater statistical robustness.